Search

Multimodel assembling methods have been commonly used to improve the reliability of climate projections by extracting relevant information from a large future possible climate scenarios. In this aspect, this study adopted the Reliability Ensemble Averaging (REA) approach to combine the precipitation outputs of multiple climate models over South Korea. The quality of REA weights assigned to each climate model was investigated using the Taylor diagram; both of these approaches showed a similar weight assignation mechanism. Furthermore, the REA performance was evaluated with the bias and root mean square error in view of reproducing the historical climate characteristics over the study region; both performance indices revealed that the REA substantially improved the performances of individual climate models at all weather stations. The analysis also showed that the climate models performance was not consistent in reproducing the precipitation characteristics over different seasons in South Korea. Thus, this study used the seasonal REA weights of each climate model for the precipitation projection. There is a general consensus between the individual climate simulators on the increasing trend of the projected precipitation in 2070–2099, except for a slight decreasing trend observed with IPSL-CM5A-LR over some parts of the central region. In general, the projected precipitation obtained with the REA over South Korea will vary between 5.23 and 10.78% for 2070–2100 relative to 1976–2005. The maximum and minimum projected precipitation increases were observed over the western and eastern parts of South Korea, respectively. [ABSTRACT FROM AUTHOR]

Multimodel combining approaches can extract reliable climate information from a large number of climate projections by exploiting the strengths and discounting the weaknesses of each climate simulator; however, most of them (e.g., reliability ensemble averaging [REA]) assign weights to climate simulators without accounting for spatial and temporal variabilities in climate model skills. Here we tested several REAs and proposed a full version that reflects the spatiotemporal (ST) variability of climate model skills. Interperformance evaluations between REA versions showed that, on average, ST‐REA reduced the bias by 33.78% and the root mean square error by 11.61%. Therefore, spatial and temporal variabilities in climate model skills can enhance the overall reliability of precipitation projections. ST‐REA was applied to project future precipitations over South Korea by combining seven climate models: The spatially averaged projected changes were 2.77%, 8.15%, and 7.58% for the 2020–2039, 2040–2069, and 2070–2099 periods, respectively. Plain Language Summary: The spatiotemporal variability of climate characteristics is one of the most pronounced climate behaviors. Thus, considering spatiotemporal variabilities during the assessment of climate models' skills when performing a multimodel averaging may improve the overall reliability of the multimodel ensemble estimate. Nevertheless, most of the available multimodel averaging approaches (e.g., the REA) do not consider in parallel the spatial and temporal variabilities during climate model weighting. In this study, we presented an improved version of the REA, which can characterize the spatiotemporal variabilities at multiple sites and time steps during multimodel averaging. The proposed REA versions have improved weight assignation mechanisms (in respect to each climate simulator) and can handle multiple weather stations. Key Points: Three augmented versions of reliability ensemble averaging were proposed to achieve reliable precipitation projections over South KoreaThe spatiotemporal variability of climate model skills within a multimodel approach improved the overall reliability of precipitation projectionsThe spatiotemporal reliability ensemble averaging outperformed all versions of reliability ensemble averaging in precipitation projection [ABSTRACT FROM AUTHOR]